DeepSeek V4.1 Flash 1017 Tops Open Cyber Index at 57%, Scores 45 on Intelligence Index
DeepSeek released DeepSeek V4.1 Flash 1017, an update to its flagship open-weights model, which scored 45 on the Artificial Analysis Intelligence Index — a 6-point gain — tying GLM-5.3 as the #2 open-weights model. It also became the top openly-available model on the Artificial Analysis Cyber Index at 57%, improving 16 points over its predecessor, and tied GLM-5.3 Flash on the ToneBench writing benchmark. The model is priced at $0.30 per 1M input tokens and $1.20 per 1M output tokens, with a 1M-token context window and MIT license.
Reference imageEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
Kimi K3.1 Pro tops openly-available models on Artificial Analysis Cyber Index at 64%
Kimi K3.1 Pro from Moonshot has become the top openly-available model for agentic cyber defense on the Artificial Analysis Cyber Index, scoring 64%, seven points ahead of the previous leading openly-available model DeepSeek V4.1 Flash 1017 (57%) and five points behind the trusted-access GPT-6 Sol (Daybreak Blue, max, 69%). Once its weights are released, K3.1 Pro will also lead open weights models for cyber defense. The index measures agentic cyber defense tasks via three benchmarks: CWE-Bench-AA, DeepsecBench-AA and CyberGym-E2E-AA, primarily through public APIs. K3.1 Pro scores 95% on CyberGym-E2E-AA, 37% on DeepsecBench-AA and 61% on CWE-Bench-AA. Its Cost per Task is $2.81, higher than GPT-6 Sol (Daybreak Blue, max, $1.77) and DeepSeek V4.1 Flash 1017 ($0.42), despite being 33% cheaper per token than Kimi K3, because it uses about twice the input tokens.
Read sourceDeepSeek V4.1 Flash 1017 Tops Openly-Available Cyber Index at 57%
DeepSeek V4.1 Flash 1017, an update to DeepSeek's flagship open-weights model, is now the top openly-available model on the Artificial Analysis Cyber Index with a score of 57%, according to Artificial Analysis. The index measures agentic cyber defense tasks across CWE-Bench-AA, DeepsecBench-AA, and CyberGym-E2E-AA, evaluated mainly through public APIs. DeepSeek V4.1 Flash 1017 ranks #2 overall, 12 points behind the trusted-access model GPT-6 Sol (Daybreak Blue, max) at 69%, and one point ahead of Grok 4.7 (xhigh) and MiMo-V2.6-Pro. It improves 16 points over its predecessor, driven by a 53-point gain on CyberGym-E2E-AA from 23% to 76%. At $0.42 per task, it sits on the capability-versus-cost Pareto frontier, over 2x MiMo-V2.6-Pro's $0.18 but 4x cheaper than GPT-6 Sol's $1.77.
DeepSeek V4.1 Flash 1017 Ties GLM-5.3 Flash on ToneBench Writing Benchmark
DeepSeek's V4.1 Flash 1017 model now ties GLM-5.3 Flash on ToneBench, a writing benchmark, according to a post from Whats_AI. Compared with the September V4.1 Flash at the same list price, the new version gains 97 Elo on high while being 21% cheaper, and matches the old version on max for $0.0095 per script instead of $0.017. On max, quality is the same within noise and 18% cheaper, using 17K thinking tokens per script instead of 22K for scripts of the same length. Against GLM-5.3 Flash (max) from Zai_org, the result is a tie, ranked #23 versus #24 of 134 configs with overlapping Elo ranges. DeepSeek costs $0.014 per script instead of $0.022 and thinks less than half as long, 17K versus 37K thinking tokens, with a script taking 1.3 minutes instead of 5.2. The poster says DeepSeek is improving the Pareto curve on writing and recommends switching from the September version or from the GLM-5.3 Flash pick.
Read sourceShow 2 older updatesHide older updates
DeepSeek V4.1 Flash 1017 Scores 45 on Intelligence Index, Ties for #2 Open-Weights Model
DeepSeek released DeepSeek V4.1 Flash 1017, an update to its flagship DeepSeek V4.1 Flash model, scoring 45 on the Artificial Analysis Intelligence Index, a 6-point improvement, making it the equal #2 open-weights model alongside GLM-5.3. The model gains in agentic capabilities, knowledge, and scientific reasoning: its GDPval-AA v2.1 Elo rating rose 146 points to 1746, AA-Briefcase v1.1 rose 150 points to 1570, AA-Omniscience non-hallucination rate improved from 4% to 33% with accuracy up from 43% to 46%, and CritPt rose from 14% to 24%. Priced at $0.30 per 1M input tokens and $1.20 per 1M output tokens during peak hours, with off-peak rates halved, it has a Cost per Task of $0.26. It has a 1M-token context window, 552B total parameters (16B active), an MIT license, and is available via DeepSeek's first-party API and providers including Fireworks AI, Baseten, DeepInfra, Wafer, and Cloudflare.
DeepSeek Releases DeepSeek-V4.1-Flash-1017 Open-Weight Model Scoring 77% on DeepSWE 1.1
DeepSeek released DeepSeek-V4.1-Flash-1017, described as a new open-weight model with frontier performance. According to the announcement, the model scores 77% on the DeepSWE 1.1 benchmark, outperforming GLM-5.3, Kimi K3, and models such as GPT-5.6 Sol and Opus 5. The model is now available on DeepSeek Chat and APIs as well as HuggingFace. The post also states that it is extremely cheap. No independent verification, methodology details, or pricing figures were provided in the source item.
Read source